13.1 Introduction

A central theme throughout earlier chapters is that the information in a SAR image is embodied in image structure (objects) together with a statistical description of the properties of these objects. Our major drive was to derive, test, and exploit progressively more highly developed data models and consequently find optimal methods to infer this structure and to characterize the objects. The endpoint of most of the work described up to now is an estimate of the relevant model parameters at each pixel based either on image reconstruction or in the context of an image segmentation. Arriving at this description of the image relies on two types of knowledge about the world: (1) semiempirical inferences about the types of statistical data likely to be needed in describing the observations from distributed targets and (2) general descriptors of scene structure carried in such concepts as the cartoon model (Chapter 7). Essentially we were concerned with measurement: what should be measured in a SAR image, which methods should be used, and how accurate are the results? At a fundamental level, these measurements are all we have. A theory of measurement that is complete tells us everything there is to know about a pixel or object in the scene and circumscribes what needs to be explained about it. Nonetheless, to make measurements useful, we need to assign high-level meanings to them, which requires considerably more world knowledge. A ubiquitous requirement is one of recognition of objects in the scene.